Caching, limits and speed

Why a report is instant the second time, what the 20,000-item cap does, and how to make a slow report fast.

The cache

Results are kept per person for 30 minutes by default. Within that window, opening the report again — or opening a dashboard full of gadgets — costs nothing and is instant.

Every report says how old its results are: "Last refreshed just now", "Last refreshed 20 min ago". Refresh asks Jira again. Within a day of the last full read it asks only for what changed since, which is cheaper; a full read happens at least daily.

A Jira admin can change the window, from 5 minutes to 4 hours, and can clear every cached result at once.

Updates between full loads

When a cached result is stale but less than a day old, RollGrid does not start again. It asks Jira what has changed since the last read and updates those work items, which is far cheaper than a full load. A five-minute overlap is deliberately re-read each time, so nothing is missed to clock drift between Jira's servers and ours.

The 20,000 limit

A report reads up to 20,000 work items. If the source holds more, RollGrid tells you, in plain words, before spending the time — and tells you again on the report and in any export.

It never quietly reads a subset. A roll-up missing a thousand children reads low, and a number that is low for an invisible reason is worse than no number.

Admins can raise the limit, but the honest fix is usually a narrower source.

Jira's rate limit

Jira limits how much an app may read per hour, and that allowance is shared. RollGrid keeps a running count and stops before it exhausts it, saying so rather than failing strangely. Admins set the hourly budget and can watch the counters on the Health tab.

Making a slow report fast

  1. Narrow the source. By far the biggest lever. A JQL with a date bound beats a whole space.
  2. Turn off "Done work items" if the report is about current work.
  3. Remove tabs you do not read. A Timesheet tab, or the "no time logged" warning, makes the report load worklogs; without them, it does not.
  4. Lengthen the cache window if the data does not need to be minutes-fresh.
  5. Use slices rather than several near-identical reports: one definition to keep up, and each slice is read only when someone opens it.